Upload
others
View
4
Download
0
Embed Size (px)
Citation preview
The impact of the European Union Emission Trading
Scheme on electricity generation sectors
Djamel KIRAT� and Ibrahim AHAMADAy z
April 16, 2009
Abstract
In order to comply with their commitments under the Kyoto Protocol, France and Germany par-
ticipate to the European Union Emission Trading Scheme (EU ETS) which concerns predominantly
electricity generation sectors. In this paper we seek to know if the EU ETS gives appropriate economic
incentives for an e¢ cient and strong system in line with Kyoto commitments. If so, electricity produc-
ers in these countries should include the price of carbon in their costs functions. After identifying the
di¤erent sub periods of the EU ETS during its pilot phase (2005-2007), we model the prices of various
electricity contracts and look at their volatilities around their fundamentals while evaluating the cor-
relation between the electricity prices in the two countries. We �nd that electricity producers in both
countries were constrained to include the carbon price in their cost functions during the �rst two years of
the operation of the EU ETS. During that period, German electricity producers were more constrained
than their French counterparts and the inclusion of the carbon price in the cost function of electricity
generation has been so much more stable in Germany than in France. Furthermore, the European market
for emission allowances has increased the market power of the historical French electricity producer and
has greatly contributed to the partial alignment of the wholesale price of electricity in France with those
in Germany.
.
Keywords: Carbon Emission Trading, Multivariate GARCH models, Structural break, Non Parametric
Approach, Energy prices.
JEL classi�cation: C14 C32 C51 Q49 Q58
�Centre d�Economie de la Sorbonne, Paris school of Economics, University Paris1 Pantheon-Sorbonne. Ad-
dress: 106-112 boulevard de l�hôpital 75013 Paris, France. Phone : 33 1 44 07 82 13. Email: djamel.kirat@univ-
paris1.fr.yCentre d�Economie de la Sorbonne, Paris school of Economics, University Paris1 Pantheon-Sorbonne.
Address: 106-112 boulevard de l�hôpital 75013 Paris, France. Phone : 33 1 44 07 82 08. Email: ahamada@univ-
paris1.fr.zThe authors would like to thank Katheline Schubert and participants of the meeting �Environmental economics" held in
Lille on February 10th, 2009 for their helpful discussions and comments. This paper will be presented at the 2009 International
Energy Workshop meeting (Venice, June 17th - 19th).
1
1 Introduction
For the implementation of the Kyoto Protocol whose objectives are in force since January 2008, the European
authorities have organized a European market for CO2 permits in January, 2005. It�s the European Union
Emissions Trading Scheme (EU ETS). It concerns mainly energy1 and industrial sectors major emitters. The
market is based on a mechanism of �cap and trade�where the actors are free receivers of annual emission
permits of CO2 at the beginning of the year. They have to achieve their commitment by providing so
many permits as tons of emitted CO2 at the end of the year. Those that have emitted more CO2 than
their allocation have to comply buying permits on the market. The energy sector and mainly the sector of
electricity generation is by far, the most CO2 emitter. Hence it has been the largest share of the Community
allocation for the period 2005-2007. What so lets glimpse narrow relations between the electricity market,
the markets of the fossil fuels used in the electricity generation and the European market for CO2 permits.
The main objective of the EU ETS is to encourage the industry�s most emitters to reduce their carbon
emissions and invest in clean technologies. So, achieving this objective is conditioned by the emergence
of a real carbon price signal. The latter would require electricity producers to make long-term choices to
produce electricity with fewer emissions. In this context, the ex-post empirical analysis of the impact of the
introduction of the European market for CO2 permits on energy markets and, particularly that of electricity,
is essential to assessing the e¢ ciency and the consequences of the introduction of EU ETS.
The electricity price is determined by the costs of fossil fuels, the impact of environmental policies and
measures and climatic factors such as temperature and raining. In Europe, it is widely agreed that gas and
coal prices account for the variations of the electricity price. Besides, economic theory teaches us that carbon
price is a marginal cost and that the carbon permit has an opportunity cost equal to its market price. It
suggests that carbon price should be included in the price of electricity. Empirically, the rough fall in the
price of CO2 of about 10 e / ton in April, 2006 which was followed at once by a movement of 5 to 10 e /
MWh on the electricity market (Reinaud, 2007) and the English company British Energy which lost 5 % of
its market capitalization in three days during the same period (Bunn and Fezzi, 2007) are all evidence of the
in�uence of the carbon market in the electricity market. Many studies have dealt with the impact of carbon
price on electricity prices of various European markets for the last three years. So Sijm et al. (2005, 2006)
have studied the case of Dutch and German electricity markets to determine the share of carbon price which is
re�ected in the price of electricity. Their study was based on an Ordinary Least Squares (OLS) estimate of a
basic linear model. Honkatukia et al. (2006) have studied the long-run and short-run dynamics of electricity
prices, gas and coal prices and the permit of carbon in the Finnish market, based on a VAR analysis. Bunn
and Fezzi (2007) adopted a similar approach to analyze the English electricity market excluding the price of
coal and including the temperature and dummies as exogenous variables. They realized a structural analysis
of the relations between energy prices and carbon price through short-run restrictions. The results of these
studies are all the more contrasting that approaches are di¤erent and the countries surveyed are of great
1 It is the oil re�ning, electricity production, heating and transporting gas.
2
diversity in their energy mixes. So, the absence of unanimous response to the problem of the e¤ect of the
EU ETS on the price of electricity (Reinaud, 2007) is mainly due to the coexistence of various electricity
markets in Europe and the heterogeneity of energy mixes of the European Union countries. Furthermore,
these studies have covered, at most, the period from January 2005 to May 2006.
This article aims at providing a clear answer about the impact of the introduction of the EU ETS on the
electricity generation sector by taking into account this heterogeneity. It deals with the volatility of electricity
price around its fundamentals and compares two European countries with very di¤erent energy mixes that are
France and Germany. we estimate a model based on the cost function of electricity generation, which includes
the cost of carbon and measures the instantaneous correlation between the wholesale electricity prices in
both countries. This paper covers all the pilot phase of the EU ETS (2005-2007) and takes into account
its di¤erent sub periods. It is organized as follows: Section 2 presents the organization and functioning of
the sector of electricity ; Section 3 is devoted to the presentation of the mechanism of the price formation
of emission permit and its impact on the sector of electricity ; Section 4 presents a descriptive analysis
of the relations between electricity markets on the one hand, primary energy markets and carbon market
on the other hand, and the steps of the econometric modelling ; Section 5 presents the results and their
interpretation ; Section 6 concludes this article.
2 The sector of electricity generation
The electricity sector has received nearly 55 % of the Community allocation of the pilot phase of the European
market for C02 permits. Before analyzing the impact of the introduction of carbon constraint on this sector,
it is advisable to present its organization and functioning. This sector is organized around four main areas :
production, transport, distribution and marketing. Purely �nancial activities such as brokerage and trading
over the counter or on power exchanges are added to these four market segments. Electricity generation is
the main polluting activity in this sector, it has been put in competition in the process of liberalizing the
electricity market in Europe from 1998. Electricity is produced from various primary energy sources such as
nuclear power, coal, oil, gas, hydropower, biomass, wind, solar and geothermal. The proportions of the use
of these di¤erent primary energy sources in electricity generation in a country determines its energy mix.
The latter is very di¤erent from an European country to another because of di¤erences in energy policies
and speci�c geographical and geological features of each country. Besides, electricity is not a good as another
because it is not storable, what confers on its generation sector particular characteristics which we detail in
the following.
3
2.1 The merit order between electricity generation technologies and the adap-
tation of supply to demand
Electricity demand is characterized by important �uctuations. It shows variations from an hour to another,
one day to the next in the week and a season to the other during the year. These changes require the
need for an instantaneous equilibrium between supply and demand, resulting in a continuous adaptation of
electricity supply to changes in demand. As electricity production presents very di¤erent costs according to
the used technology, pro�tability is di¤erent, depending on the choice of the primary energy used in electricity
generation. Therefore, electricity production is subject to a sequential use of production technologies which
depends on their production costs. The producers should rather start up power plants to meet the demand,
in increasing order of variable marginal costs of production. That is the concept of �merit order�between
the various technologies which integrate di¤erent sources of primary energy in electricity production. The
merit order is determined by the variable marginal cost of production which takes into account only the
variable costs (the costs of fuels and operational costs). It re�ects an order of pro�tability so that production
units with the lowest marginal costs are held �rst and foremost in plans for electricity generation. The merit
order between technologies is not �xed. The inclusion of the price of carbon allowances in cost functions of
the most polluting technologies can have an impact on the merit order among the primary energy used to
produce electricity and thus reverse the order of pro�tability. So, we determine the Switching price (Sijm et
al., 2005) which is the price of carbon for which it becomes more interesting for a producer to use gas power
plant rather than coal plant.
2.2 The pro�tability of electricity production: trade o¤between short-run and
long-run strategies
It is now clear that the choice of power production plan will depend on the merit order, but this is not the
only determiner of the choice of production plan. In fact, the producer will take into account the number of
hours of functioning necessary for the pro�tability of a given type of power plant. At this end, the producer
integrates into these choices of production the following criteria:
� The depreciation of �xed capital invested in the various types of power plants
It takes into account the facilities life duration, environmental costs of CO2 emissions and the energy
e¢ ciency of each fuel.
� The "availability" of kwh produced
A kwh which can be produced on demand (gas, oil and coal) allows full adaptation of supply to demand,
while a random kwh depending on the weather (solar, wind), does not satisfy the customer�s demand at the
right time. We must make use of complementary production means to respond customer demand, which
induces an additional cost.
4
� The number of hours of annual operation of each type of power station
The pro�tability of a type of power plant depends on the number of annual hours of its operation. While
nuclear power stations are pro�table when they run all year (this type of technology is not pro�table to use
less than 1500 hours per year), a gas plant is pro�table for a period of annual functioning from 1000 to 1500
hours.
So electricity producers make delicate calculations and are very sensitive assessment of the production
costs of di¤erent technologies while ensuring production following the demand curve in real time. In peak
periods, a number of production units are used. As demand decreases, the number of production units
decreases. This means stop and restart units depending on the level of demand. The operational features
of the production units (including start-up time, the levels of maximum and minimum production, energy
e¢ ciency) predestine power plants to a mode of continuous or discontinuous production. This justi�es
the presence of units of production of various types in the same park production. Logically, we can say
that during peak periods it�s best to mobilize units which have low �xed costs and high variable costs. In
contrast, outside peak periods, it�s preferable to use units that have low variable costs and high �xed costs
to be spread over a longer use. Figure 1 illustrates the supply curve of electricity resulting from the choice
between technologies of electricity generation using di¤erent sources of primary energy and re�ecting a long
term marginal cost in the absence of any carbon constraint.
Figure 1: Supply curve of electricity
5
2.3 Energy mixes and relations between the prices of primary energies and
electricity prices
In order to show the link between the carbon market, the electricity market and markets of primary energies,
it seems important to analyze the nature of energy mixes of European countries. This analysis justi�es,
among others, the choice of European countries to be included in the comparative analysis. We identify the
countries in which the links between the carbon and energy markets would be most likely to be strong and
choose among them countries with diverse energy mixes. Figure 2 represents the energy mixes of some major
European countries in 2004. Countries with predominantly gas or coal power plants are more likely to be
a¤ected by variations in the price of carbon, because power plants using coal or gas are emitting more CO2.
We then focus special attention to countries with energy mixes dominated by gas and coal. In the case of
Germany, more than half of the electricity is generated using coal and lignite, while France produces almost
80 % of its electricity from nuclear energy. Thus, the importance of coal and gas in power generation and
the concern to take into account the di¤erences between the energy mixes of the European Union countries
lead us to retain Germany and France for the comparative study.
Figure 2: Annual electricity production in Europe by country and type of primary energy (2004)
3 The Emissions Trading Scheme and its impact on electricity
producers
The emission permit of CO2 is a free traded good. Its price is determined by the meeting of supply and
demand on the market. But in the case of emission permits, it is necessary to make the distinction between
the short-term daily market and the long-term annual compliance for which market participants have com-
mitted themselves. Thus, the di¤erences in horizons between the daily market for emissions and the annual
6
commitment suggest phenomena of persistent shocks. Indeed, while agents record instantly shocks on a daily
basis, it remains that they can react more downstream to collected information by incorporating carbon into
their long term strategies. This can cause phenomena of persistent shocks.
Initially, the allowance market was scheduled to run in two phases (Phase 1: 2005-2007; Phase 2: 2008-
2012). On each of these two phases, each member of the European Union must accept a national allocation
plan to an annual reduction of CO2 emissions while retaining the prerogatives relating to the de�nition of
major variables such ceiling of the emissions attributed to the device, the list of plants that will be concerned
and the rules for allocating quotas to existing and new facilities. The plan is based on a percentage of emission
reductions for each installation of a country from the principle of "grandfathering". Therefore, there is
an obligation to reduce annual emissions of CO2, which is uncertain because of this very principle. Then,
through the European Union, there is a supply function of reduction of CO2 emissions (Bunn and Fezzi, 2007)
re�ecting increasing marginal costs of reducing emissions over a year. In the sector of electricity generation,
this supply function re�ects the changes that occur in the merit order curve between the primary energies
used in electricity generation. These changes depending on the energy mixes and installed productive parks
in each country, the supply function of reduction of CO2 emissions re�ects, for low costs of reducing CO2
emissions, the substitution of lignite by coal in electricity production in Germany, and for higher abatement
costs, the more expensive alternative of substituting gas for coal in electricity production in the United
Kingdom or Germany. The response of the sector of electricity to the obligation to reduce annual emissions
of CO2 is di¤erent from one EU country to another. It depends on the country�s energy mix and therefore
the prices of primary energies and the price reached by the quota of carbon. Hence, the supply function of
reduction of carbon emissions is convex, discontinuous, uncertain and variable through the year re�ecting
the costs of switching between technologies of power generation.
The agents are involved in the daily market for allowances by buying and selling permits for CO2
emissions. They make their decisions based on their forecasts Et [f (Dj)], where f is the function of emission
reduction supply and Dj the required emission reduction during phase j. These forecasts, which focus on
the annual equilibrium price of CO2, evolve continually during the year (Bunn and Fezzi, 2007). Therefore,
the fact that electricity producers that emit more CO2 than their allowances are starting to buy allowances
on the market to be in compliance, it is reasonable to predict that the price of carbon is added to fuel costs
and operational costs of electricity generation. On the other hand, due to free allocation of CO2 emission
allowances to participants at the beginning of the period and the emergence of a carbon price from the
daily market, these permits are a new liquid assets available to participants, swinging an opportunity cost
of emission permit equal to its market price (Sijm et al., 2006). Critics on the e¢ ciency of the EU ETS
as a means to reduce emissions using this argument to show that in the short term polluters are making
windfall pro�ts. If the European market for emissions proved to be long position, which was the case during
its pilot phase, and face the uncertainty inherent in the emission reduction supply function, participants able
to anticipate such a market development would make many windfall pro�ts.
7
4 Electricity price formation, database and econometric modelling
4.1 From stylized facts to the econometric model
Electricity wholesale markets in France and Germany are of oligopolistic market design. The price of electric-
ity results from the market clearing of supply and demand on power exchanges and is equal to the marginal
cost of electricity generation plus a mark-up. Due to the fact that electricity demand is inelastic, the relative
di¤erence between the price of electricity and its marginal cost of production remains constant. Thus, with
a constant mark-up rate, changes in electricity prices will re�ect the changes in cost of electricity generation
and the prices of electricity will depend directly on the marginal cost of producing electricity. Electricity de-
mand �uctuates continuously within a certain interval. Its curve meets the supply curve of electricity at one
point of this interval and achieves the electricity market equilibrium, thus determining the wholesale price
of electricity. The range within the demand �uctuate corresponds to minimum and maximum quantities of
electricity consumed at any time during the year. This interval coincides with the quantity of electricity
produced to meet demand from primary energy sources that may di¤er between countries with regard to
their diverse energy mixes and their levels of electricity demand. The marginal cost of electricity is equal
to the cost of primary energy used to produce the last unit of electricity, operating costs thereon, plus any
inclusion of carbon cost of production of that unit. The price of primary energy used to produce the last
unit of electricity is a major determinant of electricity prices. For these two countries, the primary energy
can be either gas or coal. More, depending on whether the cost of carbon is included or not in the electricity
generation function, the price of emissions of carbon dioxide will be a determinant of electricity price or on
the contrary it will have no in�uence on it.
Climatic variables such as temperature, rainfall or brightness may also be important determinants of the
price of electricity in a country. Indeed, the temperature and lighting can in�uence the demand for electricity
while rainfall may have an impact on the supply of electricity in a country for which the share of hydropower
in the energy mix is important. The importance of any of these variables is then dependent on the location
and composition of the country�s energy mix. In the two countries covered by our study, temperature is
crucial in electricity prices. It exerts a dual e¤ect on energy demand in general and particularly on that of
electricity. The relationship between electricity demand and the temperature is non-linear « V » shaped
function, as electricity demand increases for both low and high temperatures (Engle et al., 1986). To take
into account the nonlinearity of the relation between electricity price and temperature, we estimate this
function, in the cases of both countries, by the second order local polynomials method, in order to determine
the threshold for which the electricity price-temperature gradient is reversed. Hence, we de�ne two variables
of temperature for each country: the change in temperature above the threshold (Thot) and temperature
variation below the threshold (T cod)2 .
2To linearise the "V " shaped function, one has to consider that if during the intra-period variation the temperature crosses
the threshold the relationship is reversed. To overcome this problem Thott is de�ned as all the variation of the temperature that
occurs above the threshold and T codt as all the variation that occurs below. In fact, if temperatures in t and t+1 are such that
8
Starting from these stylized facts concerning the electricity price formation process, we estimate an
empirical time series model. The econometric speci�cation of the relation between the price of electricity and
its determinants expressed above will be done using a dynamic modelling because price variables in general
are functions of expectations formed by agents from their past experiences and new information they have,
in other words, past and contemporary prices. So one can write expectation expressed in contemporaneous
period of the value of future electricity prices as follows:
P electt = Et�P elec�t jZt; P elect�1 ; P
elect�2 ; :::
�= g
�Zt; P
elect�1 ; P
elect�2 ; :::
�(1)
Where Zt represents the new information available to agents in the current period, such energy prices
entering the electricity generation process, and P elect�i the past values of electricity price. Then, we opt
for an econometric model where the price of electricity is based on its past values, the prices of gas, coal
and emissions of carbon dioxide and temperature variables Thotand T cod presented previously. Let be the
following equation (2):
P elect = �0 +
pXi=1
�iPelect�i + �P
gast + �P coalt + P carbont (2)
+�1Tcodt + �2T
hott + "t
Where P yt the logarithm of the price of the commodity y at the period t. The price variables taken into
logarithm have the double advantage of reducing the variance and allow reasoning in terms of elasticity.
The number of lags p of the dependent variable to take as a regressor will be determined for each country,
minimizing the Akaike (AIC) or the Bayesian (BIC) information criterion.
4.2 Data and descriptive analysis
Our study aiming at identifying the responses of the sector of electricity to the introduction of the EU ETS,
we�ll use electricity prices from di¤erent contracts on the electricity stock exchanges3 of both countries in
e / MWh. As the market segment of intra-day contracts lacks liquidity and is only intended to answer
unforeseen punctual physical needs during the day, we will use the day-ahead and the month-ahead base
load4 electricity prices of the French and German electricity stock exchanges. These data sets and all those
used in this study are weekdays frequency and run from July 4th, 2005 to June 29th, 2007. Due to its
one is above the threshold for which the relationship between the price of electricity and the temperature is reversed and the
other below, then Thott = Threshold � temperature(t) and T codt = temp�erature(t + 1) � Threshold. See the appendices for
more details about determining the threshold.3 It is Powernext in France and EEX in Germany.4The base load price of electricity is the price on the block for 24 hours. It is an arithmetic average price of 24 hours of the
day (from 0h to 23h).
9
liquidity, the carbon spot price of the Powernext stock exchange expressed in e per ton will be used. On
the markets of the primary energies, the following price series expressed in e per MWh will be used. It
is the gas price of the month-ahead future contract traded on the Zeebrugge hub and the coal price of the
month-ahead future contract Coal CIF ARA. The variables of temperature Thot and T cod were built from
the Powernext daily index of temperature (expressed in degrees Celsius) for both countries. These indexes
are calculated from a weighted average, by regional population, of temperatures recorded at representative
regional weather stations of each country. Finally, we have all of a sample of 520 observations for each series
of data and we will now present their main characteristics.
Figures 3 to 5 present the evolution of prices of various electricity contracts proposed on the French and
German electricity stock exchanges, as well as the evolution in the prices of gas, coal and carbon.
Figure 3: Month-ahead contract electricity price in France and Germany.
These �gures show that the prices of various electricity contracts have varying volatilities. Prices of
day-ahead contracts are of an extreme volatility compared to those of month-ahead contracts. The price
of coal shows no major changes and remained relatively stable within a range from 6 to 8 e /ton during
the period from July, 2005 till June, 2007 with a trend towards the upper bound of the range at the end
of the period. During the same period the price of gas shows a decreasing general trend marred by large
�uctuations, including a signi�cant increase during the winter of 2005 when gas prices rose from just under
20 e / MWh in October to over 50 e / MWh in December. The spot price of carbon has �uctuated in a
range from 20 to 30 e per ton from the launch of Powernext5 until April, 2006 when the price of emissions
of carbon dioxide fell to nearly 15 e in only 3 days. This sudden collapse of the carbon price followed the
5The French carbon stock exchange was launched on �st July 2005.
10
Figure 4: Day-ahead contract electricity prices in France and Germany.
disclosure of 2005 veri�ed emissions by the European authorities. The results revealed a net long position
of the carbon market with more allowances than actual emissions.
It was a carbon market correction which induced a signi�cant break in the series of carbon spot price. It
was likened to a structural break (Alberola et al., 2008) in the sense that, the pace of the series of carbon
spot price has completely changed after the shock. Then, with the approach of the end of the pilot phase
of the EU ETS, the carbon spot price continued to decline and converge towards zero con�rming the long
position of the carbon market not only for the �rst two years of its operation but over the whole pilot phase.
In addition, the youth of the market for emission permits, source of instability of this market because players
are in learning period, suggests the presence of other structural breaks in the series of spot price of carbon.
Moreover, Alberola et al. (2008) have identi�ed two structural breaks in this series. The �rst matches in
April 2006 commented above and the other occurred on October 26, 2006 following the announcement of
a nearly 15% reduction in allowances for the second phase of the EU ETS (2008-2012). Hence, as done
by Alberola et al. (2008), we apply a unit root test with two structural breaks to detect dates of breaks
occurred on the series of carbon spot price. We choose the unit root test with double change in the mean
by Clemente Montanès and Reyes6 (1998). This test makes the dates of breaks endogenous. It includes
two test procedures each depending on detrending or not the series before performing the unit root test.
Thus, the procedure applying a �lter before the test is called AO (Additive Outlier) and serves to capture
sudden changes in the series. The one which detrends and performs the test at the same time is called IO
(Innovational Outlier) and serves to capture incremental changes in the mean of the series. The test �ndings
concerning dates of breaks are summarized in �gure 6.
The results of the test run on the logarithm spot price series of emission allowances suggest two structural
6 see the appendices for more details about this test.
11
Figure 5: Carbon, gaz and coal prices
breaks. Indeed, if we retain the results of the test by Clemente Montanès and Reyes based on the IO
procedure, we note that it detects two structural breaks that occurred on April 21, 2006 and December
28, 2006. The �rst date corresponds to a sharp drop in the price of carbon due to a market correction
following the publication by the European authorities of the results of 2005 veri�ed emissions. The second
date corresponds to the beginning of the convergence towards zero of the carbon spot price following a
change in agent�s decisions on the carbon market. Indeed, due to the mild winter in 2006 which ensured a
weaker electricity demand than in winter of 2005, agents have revised downward their forecasts concerning
the equilibrium price of carbon. The long position of the carbon market in 2005, despite a cold winter, has
prompted agents to anticipate a long position in the market in 2006 due to new information which they have
concerning the mild winter of 2006. Thus, from December 2006, the participants to the carbon market have
anticipated an exceeding in the allowances for 2006 and for the whole pilot phase of the EU ETS, including
therefore the year 2007. This induced an excess of allowance supply on the market which led to a fall in the
carbon price initiating a convergence towards zero on January 2007. These changes in agent�s expectations
have been largely in�uenced by the Stern7 review on the economics of climate change and the United Nations
conference on climate change of Nairobi, which began recalling the excess supply of allowances on European
carbon market during its �rst period of operation. Then the carbon spot price was lower than 1e per ton
in February 2007. The two structural breaks occurred on the series of carbon spot price and its convergence
7The Stern Review on the Economics of Climate Change is a report released on October 30, 2006 by economist Lord Stern
for the British government, which discusses the e¤ect of climate change on the world economy. It was heavily discussed and it
predict that the total allowances in �rst period of the EU ETS will be only 1% below projected �business as usual� emissions.
12
Figure 6: Detection of dates of structural breaks occured on log carbon Spot price series.
towards zero in the �rst period of the EU ETS were the consequences of an excess of allowances at the
beginning of period compared to actual emissions and of the lack of allowance banking from one year to
another and especially from the �rst period of the EU ETS to the second one.
4.3 Estimation of the models
In order to select the most appropriate representation to the modelling of each electricity price series, we
estimate model (2) by Feasible least squares (FGLS) for each of them. We shall retain the most relevant
models regarding the Akaike (AIC) and the Schwarz (BIC) information criterion, criteria MSE and R2
which are indicators of the explanatory power of a model. However, despite having used a robust estimation
method to Heteroskedasticity, we focus special attention to the structure of regression residuals to ensure
their good properties all the more as the price series we have are of high frequency. Concerning this last
point, since Engle (1982) we know that in the context of time series models for macroeconomic and �nancial
data, variances of the disturbances were less stable than it is generally assumed and they often varied
over time. We will then test the presence of ARCH e¤ects, a very common form of Heteroskedasticity
in the time series of high frequency. We call models (a), (b), (c) and (d) models from the equation (2)
when the series of electricity prices taken into account are respectively those of the French month-ahead
contract, the French day-ahead contract, the German month-ahead contract and the German day-ahead
contract. ARCH tests applied to the residuals of models (a) and (c) concerning the month-ahead electricity
contracts do not reject the null hypothesis of no ARCH e¤ects, while the residuals of models (b) and (d)
concerning day-ahead electricity contracts reject the null hypothesis of no ARCH e¤ects. These last two
models present an ARCH heteroskedasticity in the residuals. This presence of ARCH e¤ects requires their
modelling alongside the mean equations. In addition, studying the stability of residuals is supplemented
13
by the analysis of correlograms and partial correlograms of the disturbances. This analysis con�rms the
stability of residuals of the model (a) while correlograms of the residuals of models (c) and (d) show the
presence of a seasonality of order 5 in the prices of German electricity contracts. This seasonality is daily
during the week, because the data are often weekdays frequency. To capture the seasonality we build �ve
dummies seasoni; i = 1; 2; 3; 4; 5, each corresponding to one business day of the week j, j = Monday, ...
Friday. This gives the variable season1 which is so:
season1 =
�1 if j = monday
0 otherwiseand so on for every business day
Once the remaining seasonal variables are built, we re-estimate models (c) and (d) including variables
seasoni for i = 1; 2; ::; 5, then we check the stability of new estimated residuals. We call respectively models
(c�) and (d�) these two new models. The results show that the explanatory powers of both models and all the
selection criteria of models were signi�cantly improved. In addition, ARCH tests conclude to the presence
of ARCH e¤ects in the residuals of model (d�) but not in those of model (c�).
The presence of ARCH e¤ects in the models (b) and (d�) requires their inclusion in modelling. Thus,
models of di¤erent series of electricity prices may vary depending on whether we model ARCH or GARCH
e¤ects detected in the disturbances of preliminary regressions or take into account the existence of seasonality
as in the case of models (c) and (d). These many considerations justify the selection of following models for
modelling each electricity price series. In the case of month-ahead contracts, the prior model (a) will be used
for the �nal modelling of the French electricity price with its high explanatory power, led by both an R2 of
97% and root mean squared error (RMSE) of only 4%, and the stability of its residuals. Besides, due to its
good statistical properties, we retain the model (c�) for the �nal modelling of the price of German month-
ahead electricity contract. Indeed, the analysis of correlogram and partial correlogram of the residuals of
this model suggests that the disturbances have a white noise structure. In addition, this model presents an
R2 of 85; 7% and a root mean squared error of only 11%. Concerning the price series of day-ahead contracts,
the concern to take into account any interdependence8 of the French and German electricity markets leads
us to retain the following model with Dynamic Conditional Correlation (Engle, 2002; Engle and Sheppard,
2001) DCCE(1,1) errors:8>>>>>>>>>>>>>>><>>>>>>>>>>>>>>>:
Pelec=frat = �0 +
P2i=1 �iP
elec=frat�i + �P gast + �P coalt + P carbont
+�1Tcod=frat + �2T
hot=frat + "frat
Pelec=gert = �0 +
P2i=1 �iP
elec=gert�i + �P gast + �P coalt + P carbont
+�1Tcod=gert + �2T
hot=gert +
P5j=2 jseasonj + "
gert
("frat ; "gert ) N(0;Ht)
8The estimate results of models of the prices of month-ahead electricity contracts in both countries using SUR method are
non conclusive.
14
The model DCCE(1; 1) is de�ned as:8>>><>>>:Ht = DtRtDt
Dt = diag(ph11t;
ph22t)
Rt = (diag Qt)1=2 Qt (diag Qt)
�1=2
Where the 2� 2 symmetric positive de�nite matrix Qt is given by:
Qt = (1� �1 � �2)Q+ �1ut�1uTt�1 + �2Qt�1
With u the matrix of standardized residuals. Q is the 2 � 2 unconditional variance matrix of ut, and
�1 and �2 are non-negative parameters satisfying �1 + �2 < 1. The approach to estimate the DCC(1; 1)
model includes two steps9 . First, the conditional variance of the price of day-ahead electricity contract
in each country is estimated from a GARCH(1; 1) speci�cation at the same time as the conditional mean
equation. Thereafter, the standardized residuals of regressions performed in the �rst step are used to model
the correlation in an autoregressive way to obtain the conditional correlation matrix varying over time. The
conditional variance-covariance matrix Ht is the product of the diagonal matrix of conditional standard
deviation Dt with the conditional correlation matrix Rt and the diagonal matrix of conditional standard
deviation Dt. The Rt =
0@ 1 �12t
�21t 1
1A matrix measures the instantaneous conditional correlation between
electricity prices of day-ahead contracts on German and French power exchanges. The results of estimates
of these models are reported in Table 1 and �gure 7.
However, structural breaks in the carbon spot price series, occurred on April 21, 2006 and December 28,
2006 detected with the Clemente Montanès Reyes test using the IO10 procedure, seemed likely to have an
impact on long-run relationship between the price of carbon and those of electricity and its fundamentals.
Indeed, we have identi�ed a long-run relationship between the price of electricity and its fundamentals
based on the average correlations between these variables over the whole period from July 4, 2005 to June
29, 2007. However, the correction occurred on the carbon market after the announcement of the 2005�s
compliance results and especially the convergence towards zero of spot carbon price started on December
28, 2006 could alter this relationship by drastically reducing the weight of carbon cost in the cost function
of electricity generation. This fall in the cost of carbon, which could bring about changes in the merit order
between technologies of electricity generation, are likely to alter the long run relationship between the price of
electricity, the prices of fossil fuels and the price of carbon. To evaluate the potential impact of the structural
breaks of carbon spot price on the long-run equilibrium relationship between the prices of di¤erent electricity
contracts and their fundamentals, we proceed to test the stability of estimated coe¢ cients of carbon, gas and
coal price variables. We assume that all the other estimated coe¢ cients are stable. So we test the equality9See the annex for more details about estimating this model.10The dates of structural breaks of the carbon Spot price series detected with the AO procedure are close to the dates detected
with the IO procedure. However, the results concerning the impact of structural breaks detected with the AO procedure on
the long-run relationship between electricity prices and their determinants are less conclusive.
15
of these coe¢ cients for the periods before and after the carbon spot price structural breaks. In practice, we
test �rst the equality of these coe¢ cients for the periods before and after December 28, 2006. Then we test
the equality of coe¢ cients for the periods before and after April 21, 2006 restricting the total sample to the
period from July 4, 2005 to December 27, 2006 in order to purge our estimates of the weight of any changes
occurring after December 28, 2006. The results of these stability tests suggest that the long-run relationship
between the price of electricity, the fossil fuels prices and the carbon spot price is unstable over the whole
period and that it has changed from December 28, 2006 in the cases of the German electricity contracts
and the French day-ahead one. Furthermore, these results support that, for all electricity contracts, the
correction occurred on the carbon market in April 2006 did not a¤ect the long-term relationships.
The carbon spot price structural break occurred on December 28, 2006 a¤ected the long-run equilibrium
relationship between the price of electricity and its fundamentals in the cases of German electricity contracts
and the French day-ahead one. This justi�es the estimation of models of these electricity contracts over two
sub-periods: the period from July 04, 2005 to December 27, 2006 and the period from December 28, 2006
to June 29, 2007 for the purpose comparisons. Then, the next section is devoted to interpret the estimation
results.
5 Results and interpretation
Table 1 presents the estimation results of the models used for modelling the prices of various French and
German electricity contracts over the whole period from July 4, 2005 to June 29, 2007 and by sub-periods.
We focus �rst on full period results and, in a second step, we comment the results by sub periods.
Over the whole period, all the estimated coe¢ cients at 5% level of signi�cance have the expected signs.
The estimated parameters of the logarithmic price variables in the mean equations are interpreted as long-run
elasticities because the models re�ect the long-run relationships. Higher values of estimated coe¢ cients of
lagged electricity price variables and their degrees of signi�cance re�ect the high dependence of contemporary
electricity prices on those of the previous periods. This dependence is due to expectations of contemporary
electricity prices made by agents in previous periods. These results argue that temperatures do not a¤ect
the prices of month-ahead electricity contracts while the estimated coe¢ cients of temperature variables Thot
and T cod re�ect the fact that overall, softening temperatures in�uence the price of day-ahead electricity
contracts to decline, and that variations in temperatures towards extreme values in�uence these prices in
increase. In particular, a positive variation of temperatures above the threshold, all things being equal,
leads to higher prices of French and German day-ahead electricity contracts, in the same proportions, while
a positive change in temperatures below threshold leads to a decrease in prices in di¤erent proportions. In
the latter case, higher temperatures below the threshold in France and Germany, in the same proportions,
will cause a decline in the price of the French electricity contract twice as large as in the German one.
As the temperature a¤ects the price of electricity only through electricity demand, we easily justify that
16
Table 1. Estimation results of the selected models
Country
Contract
France
Month
ahead
France
Day
Ahead
Germany
Day
Ahead
Germany
Month
Ahead
Periodfull
period
full
period
before
break
after
break
full
period
before
break
after
break
full
period
before
break
after
break
Mean equation
P elect�1 0.937*** 0.610*** 0.589*** 0.565*** 0.471*** 0.443*** 0.357*** 0.656*** 0.773*** 0.428***
(0.015) (0.052) (0.064) (0.090) (0.058) (0.058) (0.129) (0.13) 0.086 (0.121)
P elect�2 0.214*** 0.215*** 0.159 0.251*** 0.246*** 0.212** 0.202 0.122 0.078
(0.051) (0.065) (0.097) (0.056) (0.071) (0.106) (0.129) 0.080 (0.152)
P gast 0.043*** 0.094*** 0.125*** 0.040 0.128*** 0.170*** 0.070 0.080*** 0.062*** -0.030
(0.010) (0.030) (0.036) (0.097) (0.042) (0.043) (0.115) (0.026) 0.023 (0.132)
P coalt 0.007 -0.181* -0.017 0.638 0.057 0.140 1.49* 0.068 0.056 0.541
(0.026) (0.104) (0.145) (0.613) (0.145) (0.146) (0.904) (0.076) 0.073 (0.367)
P carbont -0.001 0.004 0.067*** 0.019 0.019** 0.093*** 0.016 0.002 0.022** -0.019
(0.002) (0.007) (0.023) (0.022) (0.009) (0.021) (0.027) (0.004) 0.011 (0.019)
Thot -0.000 0.012*** 0.006 0.028*** 0.011*** 0.007 0.031*** 0.004* 0.005* 0.001
(0.001) (0.004) (0.005) (0.008) (0.004) (0.005) (0.008) (0.002) 0.003 (0.004)
T cod 0.001 -0.031*** -0.030*** -0.037*** -0.014*** -0.012** -0.028** -0.013 -0.009** 0.086
(0.001) (0.004) (0.006) (0.007) (0.004) (0.005) (0.013) (0.008) 0.004 (0.055)
cons 0.103* 0.732*** 0.181 -0.431 0.695** 0.313 -1.52 0.084 -0.027 0.582
(0.057) (0.222) (0.344) (1.230) (0.329) (0.350) (1.83) (0.151) 0.164 (0.883)
season2 -0.076** -0.070* -0.067 0.083*** 0.076*** 0.086
season3 -0.193*** -0.176*** -0.174** 0.105*** 0.084*** 0.126**
season4 -0.240*** -0.217*** -0.251*** 0.106*** 0.097*** 0.116*
season5 -0.406*** -0.397*** -0.332*** 0.089*** 0.073*** 0.117*
Conditional variance equation
cons 0.001*** 0.002*** 0.01 0.005*** 0.04*** 0.011*
ARCH 0.173*** 0.159*** 0.155** 0.251*** 0.244*** 0.584**
GARCH 0.813*** 0.809*** 0.804*** 0.665*** 0.675*** 0.352*
likeliho 954.13 164.69 111.28 59.83 135.19 139.33 14.68 376.73 431.06 45.58
AIC -1894.26 -307.39 -200.56 -97.66 -240.39 -248.67 0.633 -729.46 -838.12 -67.17
BIC -1864.50 -260.64 -156.99 -65.95 -176.64 -189.25 43.87 -678.46 -790.65 -32.58
Standard errors are in () ; * ** and *** refer respectively to the 10%, 5% and 1% signi�cance levels of
estimated coe¢ cients.
17
temperature in�uences only the prices of day-ahead electricity contracts. Indeed, the short term of day-ahead
contracts and the di¢ culty to predict with accuracy the level of temperature beyond a few days explain that
the electricity supply intended to meet the changes in electricity demand due to temperature variations is
provided through day-ahead contracts. We note, however, that electricity prices in Germany introduce a
daily seasonality during the week. For the day-ahead contract, this seasonality is manifested by a decrease in
electricity prices during the week. So, being on Tuesday, Wednesday, Thursday or Friday results, all things
being equal, respectively, in a reduction of the logarithm of the day-ahead contract price of 7%, 19%, 24%
and 40% compared to the �rst day of the week. The falling price of electricity by 40% on Friday compared
to Monday is due to reduced demand for electricity over the weekend as the day-ahead contracts traded on
Friday to match the electricity demand of Saturday.
The estimation results of mean equations argue that there are important di¤erences between countries
and electricity contracts in the way the costs of primary energies and carbon are included in the cost function
of electricity generation. In France, only the price of gas has an impact on the price of electricity at the
long-run equilibrium. However, this impact is twice as high on the day-ahead contract compared to the
month-ahead electricity contract. Indeed, all things being equal, higher gas prices by 1% lead to an increase
of 0:04% of the month-ahead contract price and 0:09% of the day-ahead one. The price of carbon, on average
over the whole period, was not a determinant of the prices of French electricity contracts. In Germany, the
price of gas, unlike that of coal, has been a determinant of the price of both contracts of electricity. Thus,
all things being equal, higher gas prices by 1% result an increase of 0:08% of the month-ahead contract price
and 0:13% of the day-ahead one. The price of carbon had an impact only on the price of day-ahead contract.
A rise of 1% in the price of emission permit results, all things being equal, an increase of 0:02% of the price
of German day-ahead electricity contract. The elasticity of electricity price relative to gas price is higher
for day-ahead electricity contracts than month-ahead contracts. This re�ects a less important weight of gas
price in the cost function of electricity traded through month-ahead contracts compared with day-ahead ones.
This result reinforces the idea that electricity generation is subject to a merit order between technologies,
based on variable marginal costs. Gas price has a greater impact on electricity prices in Germany compared
to those in France. This di¤erence in the weight of gas prices in the cost of electricity generation is due to
the heterogeneity of the French and German energy mixes, shares of gas being respectively of 3% and 10%.
The estimates of the conditional variance equations over the whole period argue that electricity price
volatilities of German and French day-ahead contracts are variable. It is stronger for the French contract
compared to the German one. Indeed, the sum of ARCH and GARCH coe¢ cients is higher in France
than in Germany. So is the variance of electricity price around its fundamentals, mainly the price of carbon
dioxide, hence a greater stability of the price of German day-ahead electricity contract around its long run
equilibrium path compared to the price of French day-ahead one.
The impact of carbon prices on electricity prices mentioned above, represents an average impact over
the whole period from July 4, 2005 to June 29, 2007 whereas, stability test results argue that the structural
18
break of carbon spot price occurred on December 28, 2006 has a¤ected the long-run equilibrium relationship
between the prices of German day-ahead and month-ahead electricity contracts and the French day-ahead
contract on the one hand, and the prices of gas, coal and emission permits on the other. However, this
structural break has not a¤ected the stability of the estimated coe¢ cients of the model (a). We can deduce
that carbon spot price was not a determinant of electricity price of French month-ahead contract. Indeed,
these two series of prices are completely disconnected as the estimated coe¢ cient of carbon price in model
(a) is not signi�cant and structural breaks occurred on carbon spot price series did not a¤ect the estimated
coe¢ cients of this model. Moreover, the instability of estimated coe¢ cients of the other models, induced by
the structural break occurred on December 28th, 2006 on the series of carbon spot price, is an evidence of
the close link between the prices of electricity contracts of these models and the price of emission permits,
despite the non signi�cance in some models of the estimated coe¢ cient of carbon prices on average over the
whole period.
Table 1 contains also the estimation results of models over the periods before and after the structural
break of December 28, 2006 occurred on carbon spot price series. These results suggest that carbon spot
price has been a determinant of the price of electricity throughout the period before the convergence towards
zero of carbon spot price and therefore during the �rst two years of the EU ETS operation. Then, the price
of electricity was completely disconnected from the carbon spot price in the last year of the pilot phase of EU
ETS. During the period from July 4, 2005 to December 27, 2006 an increase of 1% in carbon price resulted
respectively, all things being equal, an increase by 0; 093%, 0; 067% and 0; 022% of prices of German and
French day-ahead electricity contracts and German month-ahead contract. The elasticities of the prices of
day-ahead electricity contracts relative to gas prices were higher during this period compared with the whole
period. However, coal price has not been a determinant of the price of any electricity contract. During the
period from December 28, 2006 to June 29, 2007, electricity prices of the various contracts were completely
disconnected from the spot price of carbon. During this period the price of gas was not a determinant of
electricity prices of the various contracts and the price of coal has been without e¤ect on them. This re�ects
distortion in long-run equilibrium relationships between the prices of electricity of the three contracts and
their determinants. The analysis of selecting models criteria con�rms this result. The structural break
of carbon spot price occurred on December 28, 2006 distorted the entire relationship between the price of
electricity and the prices of gas, coal and carbon permit. This may be an evidence of a change in the merit
order between technologies of electricity generation caused by the structural break in the series of carbon
spot price. This also proves that the European market for emission allowances had an impact on the sector
of power generation in both countries even if it was not of the same magnitude.
19
Figure 7: Conditional correlation of French and German electricity prices of day-ahead contracts
Figure 8: Log carbon Spot price series
Figure 7 presents the dynamic of the conditional correlation between the prices of day-ahead electricity
contracts in France and Germany. This correlation was positive and highly signi�cant. It has been stable
around 0,3 during the period before the structural break occurred in the series of carbon spot price in
20
December 2006. Then it dropped signi�cantly by almost 30%, following the convergence towards zero of the
carbon spot price (Figure 8), reaching a stable value of about 0,2. The stability of the conditional correlation
over each sub period is due to the extreme values of estimated coe¢ cients of the model DCCE(1,1). In fact,
over each sub period, b�1 ' 0 and b�2 ' 1 with b�1 + b�2 < 1, which implies Qt ' Qt�1 and therefore:0@ 1 �12t
�21t 1
1A '
0@ 1 �12t�1
�21t�1 1
1A. This result is con�rmed by a comparison test between a DCCmodel and a Constant Conditional Correlation (Nakatani and Teräsvirta, 2009; Bauwens et al., 2006) CCC
model. Electricity prices in Germany and France were much more correlated during the �rst two years of
operation of the EU ETS than over the period that followed. The highest correlation has coincided with the
period during which the electricity producers had been most constrained by the EU ETS. So, in a context
where the debate on the possible manipulation of the electricity wholesale prices in France by the historical
producer is still relevant today11 , it seems reasonable to expect that the wholesale prices of electricity in
France are partially aligned with those of Germany12 . Indeed, a positive and signi�cant correlation between
German and French electricity prices, even though the price of carbon was not signi�cant, corroborates the
idea that the French electricity producers take advantage of the French energy mix in terms of production
costs. During the �rst two years of EU ETS operation, the carbon market has allowed French electricity
producers to pull more pro�ts from the composition of their productive parks. This can be explained by the
stronger correlation between French and German electricity prices during the same period.
Finally, the comparative study of the impact of the introduction of the EU ETS on the sectors of electricity
generation of both countries will be comparing the elasticity of electricity price relative to carbon price for
each type of electricity contract. Regarding day-ahead electricity contracts, we note that during the �rst
sub-period, the elasticity of electricity price compared to the price of carbon is higher in Germany than in
France. For the month-ahead electricity contracts, we �nd that the price of carbon is not a determinant
of the price of electricity in France while the price of carbon has been a determinant of electricity price in
Germany during the �rst two years of the EU ETS. We conclude that German electricity producers have
more undergone the carbon constraint than their French counterparts. This is largely explained by the
di¤erences in composition of the energy mixes of both countries. This �nding is supported by the biggest
stability of the price of German day-ahead electricity contract around its path of long-run equilibrium during
the �rst two years of EU ETS operation, compared to the price of the French day-ahead electricity contract.
Indeed, during this period, comparing the sum of the coe¢ cients of ARCH and GARCH e¤ects of each
model representing these prices shows that a deviation of the price of day-ahead electricity contract from
11This debate has been animated by recent suspicions that hang over electricity producers in France. Indeed, in a recent press
release dated from March 11th, 2009, the European Commission suspects an illegal conduct of the French historical producer
of electricity. The suspected illegal conduct may include actions to raise prices on the French wholesale electricity market.12As highlighted by Glachant (2007), if one cannot �nd obvious sources of price manipulation in France, one can assume that
the French historical and dominant electricity producer leaves the setting wholesale prices in France to the competitive fringe.
These competitors are building gas power plants, this means that opening the market get rid of the economic e¤ects of the
French energy mix.
21
its equilibrium path following, all things being equal, an overestimation or underestimation of the price of
carbon by the electricity producers, the return to equilibrium is faster in Germany than in France.
6 Conclusion
We modelled and estimated the relationship between electricity prices, the prices of primary energies used
in electricity generation and the price of carbon dioxide emission permits, for France and Germany. This
enabled to re�ect the heterogeneity of responses in the sectors of electricity generation to carbon constraint
and evaluate the e¢ ciency of the EU ETS taking into account this heterogeneity. We have shown that
the impact of carbon constraint on electricity generation sectors, during the pilot phase of the EU ETS,
depended on the energy mix of the country. During that period, the impact experienced two phases. The
�rst one concerns the �rst two years of the EU ETS during which electricity producers included the cost
of carbon in their production cost function, the second during which the carbon constraint had no more
weighed on the decisions of electricity producers. However, producers in countries using predominantly fossil
fuels, big carbon emitters, had more undergone carbon coercion and thus more included the price of emission
permits in their cost function of electricity generation. The conditional correlation between electricity prices
of day-ahead contracts in France and Germany had dropped by 30% between the two phases. This drop
was due to the collapse of carbon price and its convergence towards zero. Hence, the EU ETS increased the
market power of the historical French electricity producer and greatly contributed to the partial alignment of
the wholesale price of electricity in France with those in Germany. Throughout the whole pilot phase (2005-
2007), the e¢ ciency of the European market for emission allowances had not been up to compel electricity
producers to reduce their emissions and invest in cleaner technologies. However, it was a good step towards
achieving the objectives of the Kyoto Protocol. The ine¢ ciency of the EU ETS was mainly due to the
largesse granted by the national authorities of European countries for their power generation sectors which
were considered as strategic on the one hand, and certain mechanisms de�ning devices of the EU ETS, on
the other. Thus, excess allocations and the impossibility of �banking" on the following periods have made
the horizon of the carbon market bounded and eventually prevented the creation of a scarcity which is the
essence of the carbon coercion. This greatly contributed to the convergence towards zero of carbon spot price
at the end of the pilot phase of the EU ETS, loosening the carbon coercion that was carried on producers
of electricity during the �rst two years of the operation of the carbon market.
22
A APPENDICES
A.1 Non parametric estimates of electricity price and temperature relation-
ships
Non parametric regression of French electricity price on temperature
Non parametric regression of German electricity price on
temperature
23
A.2 The test by Clemente Montanès and Reyes
The Clemente Montanès and Reyes (1998) test with double change in the mean (1998) using the AO proce-
dure and implemented to a series y is based on the estimation of the following equation:
yt = �+ �1DU1t + �2DU2t + eytWhere DUmt = 1 for t � Tbm and 0 otherwise, for m = 1; 2. Tb1 et Tb2 are the dates of structural breaks
and will be searched by the scan method. The noise of this equation becomes the dependent variable on the
equation to estimate follows:
eyt = kXi=1
!1iDTb1;t�i +kXi=1
!2iDTb2;t�i + �gyt�1 + kXi=1
�i�gyt�i + etWhere DTbm;t = 1 for t = Tbm + 1 and 0 otherwise for m = 1; 2. This equation is estimated for each
pair (Tb1; Tb2) in search of the least t-statistic of the unit root hypothesis that is then compared with values
tabulated by the authors. In addition, the same test applied to the series yt using the IO procedure is based
on the estimation of the following equation:
yt = �+ �1DU1t + �2DU2t + '1DTb1;t + '2DTb2;t + �yt�1 +kXi=1
�i�yt�i + et
Test the unit root hypothesis returns to test whether the coe¢ cient � is not signi�cantly less than 1.
A.3 Two-step estimation of DCCE models
The estimation of parameters of multivariate models is based on the method of maximum likelihood. So
with Gaussian residuals, the likelihood function is:
LT =TXt=1
log f(yt j �; �; It�1)
Where f(yt j �; �; It�1) = jHtj�12 g(H
� 12
t (yt��t)) the density function of yt given the vector of parameters
� and �. We assume that (yt � �t) N(0; IN ). Thus, the loglikelihood function is:
LT (�) = �1
2
TXt=1
�log jHtj+ (yt � �t)0H�1
t (yt � �t)�
The Gaussian likelihood provides a consistent quasi-likelihood estimator even if the true density is not
Gaussian. In the case of a DCC model the loglikelihood is composed of two parts. The �rst part depends on
the parameters of volatility and the second part depends on the parameters of the conditional correlations
knowing the volatility parameters. So, with Ht = DtRtDt we obtain:
LT (�) = �1
2
TXt=1
�log jDtRtDtj+ u0tR�1t ut
�
24
where ut = D�1t (yt � �t) and u
0tR
�1t ut = (yt � �t)
0D�1t R�1t D�1
t (yt � �t). With these notations, the
loglikelihood is:
LT (�) = �1
2
TXt=1
�log jDtRtDtj+ u0tR�1t ut
�
LT (�) = �12
TXt=1
[2 log jDtj+ u0tut]| {z }�1
2
TXt=1
�log jRtj+ u0tR�1t ut � u0tut
�| {z }
Q1LT (��1) Q2LT (�
�1; �
�2)
with ��1 the parameters of the conditional variance Dt and ��2 those of the conditional correlation Rt.
Then the loglikelihood function can be written as follows:
LT (�) = Q1LT (��1) +Q2LT (�
�1; �
�2)
(��1; ��2) are found in two stages. At the �rst stage we estimate �
�1 = argmaxQ1LT (�
�1) and at the second
stage we estimate ��2 = argmaxQ2LT (��1; �
�2).
References
[1] Ahamada, I., Flachaire, E., 2008. Econométrie Non-paramétrique. Economica, Paris.
[2] Alberola, E., Chevallier, J., Chèze, B., 2008. Price drivers and structural breaks in European carbon
prices 2005-2007. Energy Policy 36, 787-797.
[3] Benz, E., Trück, S., 2009. Modeling the price dynamics of CO2emission allowances. Energy Economics
31, 4-15.
[4] Bauwens, L., Laurent, S., Rombouts, V.K., 2006. Multivariate GARCH models: A survey. Journal of
Applied Econometrics 21, 79-109.
[5] Clemente, J., Montanès, A., Reyes, M., 1998. Testing for a unit root in variables with a double change
in the mean. Economics Letters 59, 175-182.
[6] Derek, W.B., Fezzi, C., 2007. Interaction of European Carbon Trading and Energy Prices. Fondazione
Eni Enrico Mattei, Working paper 123.
[7] Engle, R.F., 1982. Autoregressive Conditional Heteroscedasticity with Estimates of the Variance of
United Kingdom In�ation. Econometrica 50(4), 293-302.
[8] Engle, R.F., 2002. Dynamic Conditional Correlation: A Simple Class of Multivariate Generalized Au-
toregressive Conditional Heteroskedasticity Models. Journal of Business and Economic Statistics 20,
339-350.
25
[9] Engle, R.F., Granger, C.W.J., Rice, J., Weiss, A., 1986. Semiparametric Estimates of the Relation
Between Weather and Electricity Sales. Journal of the American Statistical Association 394, vol. 81,
310-320.
[10] Engle, R.F., Sheppard, K., 2001. Theoretical and Empirical Properties of Dynamic Conditional Corre-
lation Multivariate GARCH. Stern Finance Working Paper Series {FIN}-01-027, New York University
Stern School of Business.
[11] European Commission, March 2009. MEMO/09/104 available at: http://europa.eu/ rapid/ pressRe-
leases Action .do?reference=MEMO/09/104&format=HTML&aged=0&language=EN&guiLanguage=fr.
[12] Galli, A., Armstrong, M., 2003. Comparing the Spot Prices from Powernext and EEX. Cerna Paris
Working Paper available at: http://www.cerna.ensmp.fr/Documents/AG-MA-Powernext-EEX.pdf.
[13] Galli, A., Armstrong, M., 2005. Are day-ahead prices for electricity converging in conti-
nental Europe? An exploratory data approach. Cerna Paris Working Paper, available at:
http://www.cerna.ensmp.fr/Documents/MA-AG-PriceConvergence.pdf.
[14] Glachant, J.M., 2007. Insatisfactions françaises sur le marché européen de l�électricité. ADIS-GRJM
Working Paper.
[15] Honkatukia, J., Mälkönen, V., Perrels, A., 2006. Impacts Of The European Emission Trade System On
Finnish Wholesale Electricity Prices. VAAT Discution paper 405, Government Institute for Economic
Research, Helsinki.
[16] Kanen, J.L.M., 2006. Carbon Trading and Pricing. Environmental Finance Publication.
[17] Mansanet-Bataller, M., Pardo, A., Valor, E., 2007. CO2 prices energy and weather. The Energy Journal
28 (3), 67-86.
[18] Nakatani, T., Teräsvirta, T., 2009. Testing for Volatility Interactions in the Constant Conditional Cor-
relation GARCH Model. Econometrics Journal 12, 147-163.
[19] Reinaud, J., 2007. CO2 allowances & electricity price interaction: impact on industry�s electricity
purchasing strategies in Europe. IEA information paper.
[20] Sijm, J.P.M., Chen, Y., Bakker, S.J.A., Harmsen, H., Lise, W., 2005. CO2 price dynamics: the impli-
cations of EU emissions trading for the price of electricity. ECN report No. ECN-C�05-081.
[21] Sijm, J.P.M., Chen, Y., Donkelaar, M. ten., Hers, J.S., Scheepers, M.J.J., 2006. CO2 price dynamics:
a follow-up analysis of the implications of EU emissions trading for the price of electricity. ECN report
No. ECN-C-06-015.
26